Hi, I'm
Actuary · UNAM
I trained as an actuary at UNAM, and since then I haven't stopped building: a transformer trained on Proust, data platforms on GCP, AI agents over Mexican regulation, production APIs. Every project starts with a question I can't answer yet and ends with something that works.
Valuing life insurance reserves requires connecting mortality projection, product design, and regulatory capital in one continuous flow. SIMA handles it end-to-end: Lee-Carter mortality projection, commutation tables, reserve valuation for three products, and RCS capital requirements with stress testing under LISF. Deployed on Google Cloud.
CreditGraph shows how relationships between owners, companies, and guarantees change portfolio review. Explore four fictional scenarios, connected balances, and a separate experiment comparing models with and without network features.
An insurance claim travels a long path between the event and the model that prices it. Automating that flow produces faster, more reliable, and more consistent data. This project builds every segment on GCP: real-time ingestion with Pub/Sub and Beam, dimensional warehouse in BigQuery, Dagster orchestration, Terraform infrastructure, and a Tweedie GLM that turns clean data into actuarial premium. Six stages, one continuous flow.
Quantitative financial risk is not learned from a single model; it requires building progressively from the foundations. This series spans 13 projects ranging from calculating how much a portfolio can lose in a day (VaR) to modeling how one institution's failure spreads across the entire financial system. Covers equities, bonds, options, and systemic risk with real market data.
Language models are used daily but few people understand what happens inside. This character-level transformer, trained on all 7 volumes of In Search of Lost Time and built from scratch in PyTorch, makes that mechanism visible: how embeddings are learned, how multi-head attention operates, and why everything reduces to matrix multiplication.
Pricing major medical insurance without real claims data is guesswork. This project processes 5.1M claims and 95.9M insured-years from CNSF open data (2020-2024), classifies them into three hospitalization levels with AI, and calculates the net premium via frequency-severity adjusted for medical inflation. The output: an interactive tariff calculator on Vercel.
Interactive artifacts and technical notes on actuarial science, finance, and statistics.
The reference I built to prepare for Exam P. Covers general probability, random variables, and multivariate distributions, each topic with actuarial intuition, not just formulas. Includes solved problems with analysis of why the distractors mislead.
One document tying four worlds together: how to price what nobody knows the value of (exotic derivatives), how to read the future in Treasury yield curves, how two firms can swap risk through an IRS, and how to build a portfolio that doesn't rely on luck. All solved in Python and bound by the same no-arbitrage logic. Integrative exam for Quantitative Methods in Finance, UNAM.
Same question, two ways of thinking: the classical approach says there's not enough evidence, the Bayesian one says B is almost certainly better. This summary explores what happens when you stop relying solely on the p-value and start thinking in terms of expected utility to decide. Personal statistical exploration project. Available in both languages.
National Autonomous University of Mexico (UNAM)
Faculty of Sciences · 2021 – 2025
100% credits completed, degree in progress
